AI-Based Resource Allocation in Heterogeneous Computing Architectures

Abstract

Heterogeneous computing architectures, comprising a mix of CPUs, GPUs, FPGAs, and specialized accelerators, present significant challenges in resource allocation due to varying workloads, hardware capabilities, and energy constraints. Traditional resource management techniques often fail to optimize performance and energy efficiency in such dynamic environments. This study explores the application of artificial intelligence (AI)-based approaches for intelligent resource allocation in heterogeneous computing systems. By leveraging deep reinforcement learning (DRL), heuristic optimization algorithms, and predictive workload modeling, we propose an AI-driven framework that dynamically allocates resources based on workload characteristics, system constraints, and real-time performance metrics. Experimental evaluations demonstrate that the proposed AI-based resource allocation strategy improves execution efficiency, reduces energy consumption by up to 30.2%, and enhances overall system throughput compared to conventional load-balancing methods. The study also discusses the challenges of deploying AI in real-world heterogeneous computing environments, including scalability, adaptability, and model interpretability. The findings underscore the potential of AI in transforming resource management strategies for next-generation computing architectures.